Role of the Cerebellum in Time-Critical Goal-Oriented Behaviour: Anatomical Basis and Control Principle

Abstract

sensory stimuli.. Anatomically, the cerebellum projects to all extra-striate components of the fast sensory-motor route: Posterior parietal cortex (PPC), Premotor cortex (PM) and Motor Cortex (M1). Indirect evidences suggest that the cerebellum sets up stimulus-response (S-R) sets at the level of the PPC. Among unresolved issues is the question of how S-R mappings are activated in PPC, how planning is performed and how the cerebellum is informed of plans. Computationally, the proposed principle of off-line planning of S-R associations poses interesting problems: i) planning must now define both the stimulus S and the action R that it will trigger. ii) There is uncertainty on the stimulus that will appear at the time of execution. Hence the planning process needs to produce not a single optimal solution but a field of solutions. It is proposed here that problem i) can be solved if only learned S-R associations are involved. Problem ii) can be solved if the neural network for S-R mapping has appropriate generalization properties. This is demonstrated with an artificial neural network example using normalized radial basis functions (NRBF). Planning a single optimal trajectory enables to generate appropriate motor command even for initial states outside of the optimal trajectory. Current implementations include a simulated robot arm and the control of a real autonomous wheelchair. In terms of control theory, the principles proposed in this paper unify purely behaviour-based approaches and approaches based on planning using internal representations. On one hand sensory-motor associations enable fast reactions and on the other hand, being products of planning, these associations enable flexible goal adaptation.